Project Goal¶

To predict the coffee stock closing price using Facebook Prophet

In [1]:
%pip install prophet
import numpy as np
import pandas as pd
import seaborn as sns

import matplotlib.pyplot as plt
from plotly.subplots import make_subplots
import plotly.graph_objects as go
import plotly.offline as pyo
pyo.init_notebook_mode()

from prophet import Prophet

%matplotlib inline
import warnings
warnings.filterwarnings("ignore")

Defaulting to user installation because normal site-packages is not writeableNote: you may need to restart the kernel to use updated packages.
Requirement already satisfied: prophet in c:\users\gentb\appdata\roaming\python\python39\site-packages (1.1.1)
Requirement already satisfied: numpy>=1.15.4 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (1.21.2)
Requirement already satisfied: matplotlib>=2.0.0 in c:\programdata\anaconda3\lib\site-packages (from prophet) (3.5.1)
Requirement already satisfied: python-dateutil>=2.8.0 in c:\programdata\anaconda3\lib\site-packages (from prophet) (2.8.2)
Requirement already satisfied: holidays>=0.14.2 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (0.18)
Requirement already satisfied: pandas>=1.0.4 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (1.1.5)
Requirement already satisfied: setuptools>=42 in c:\programdata\anaconda3\lib\site-packages (from prophet) (61.2.0)
Requirement already satisfied: wheel>=0.37.0 in c:\programdata\anaconda3\lib\site-packages (from prophet) (0.37.1)
Requirement already satisfied: tqdm>=4.36.1 in c:\programdata\anaconda3\lib\site-packages (from prophet) (4.64.0)
Requirement already satisfied: convertdate>=2.1.2 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (2.4.0)

Requirement already satisfied: cmdstanpy>=1.0.4 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (1.0.8)
Requirement already satisfied: LunarCalendar>=0.0.9 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (0.0.9)
Requirement already satisfied: setuptools-git>=1.2 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from prophet) (1.2)
Requirement already satisfied: pymeeus<=1,>=0.3.13 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from convertdate>=2.1.2->prophet) (0.5.12)
Requirement already satisfied: hijri-converter in c:\users\gentb\appdata\roaming\python\python39\site-packages (from holidays>=0.14.2->prophet) (2.2.4)
Requirement already satisfied: korean-lunar-calendar in c:\users\gentb\appdata\roaming\python\python39\site-packages (from holidays>=0.14.2->prophet) (0.3.1)
Requirement already satisfied: ephem>=3.7.5.3 in c:\users\gentb\appdata\roaming\python\python39\site-packages (from LunarCalendar>=0.0.9->prophet) (4.1.4)
Requirement already satisfied: pytz in c:\programdata\anaconda3\lib\site-packages (from LunarCalendar>=0.0.9->prophet) (2021.3)
Requirement already satisfied: kiwisolver>=1.0.1 in c:\programdata\anaconda3\lib\site-packages (from matplotlib>=2.0.0->prophet) (1.3.2)
Requirement already satisfied: pyparsing>=2.2.1 in c:\programdata\anaconda3\lib\site-packages (from matplotlib>=2.0.0->prophet) (3.0.4)
Requirement already satisfied: packaging>=20.0 in c:\programdata\anaconda3\lib\site-packages (from matplotlib>=2.0.0->prophet) (21.3)
Requirement already satisfied: fonttools>=4.22.0 in c:\programdata\anaconda3\lib\site-packages (from matplotlib>=2.0.0->prophet) (4.25.0)
Requirement already satisfied: pillow>=6.2.0 in c:\programdata\anaconda3\lib\site-packages (from matplotlib>=2.0.0->prophet) (9.0.1)
Requirement already satisfied: cycler>=0.10 in c:\programdata\anaconda3\lib\site-packages (from matplotlib>=2.0.0->prophet) (0.11.0)
Requirement already satisfied: six>=1.5 in c:\programdata\anaconda3\lib\site-packages (from python-dateutil>=2.8.0->prophet) (1.16.0)
Requirement already satisfied: colorama in c:\programdata\anaconda3\lib\site-packages (from tqdm>=4.36.1->prophet) (0.4.4)
In [2]:
df = pd.read_csv("E:\coffee.csv", parse_dates=['Date'])

df.head()
Out[2]:
Date Open High Low Close Volume Currency
0 2000-01-03 122.25 124.00 116.10 116.50 6640 USD
1 2000-01-04 116.25 120.50 115.75 116.25 5492 USD
2 2000-01-05 115.00 121.00 115.00 118.60 6165 USD
3 2000-01-06 119.00 121.40 116.50 116.85 5094 USD
4 2000-01-07 117.25 117.75 113.80 114.15 6855 USD
In [3]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 5746 entries, 0 to 5745
Data columns (total 7 columns):
 #   Column    Non-Null Count  Dtype         
---  ------    --------------  -----         
 0   Date      5746 non-null   datetime64[ns]
 1   Open      5746 non-null   float64       
 2   High      5746 non-null   float64       
 3   Low       5746 non-null   float64       
 4   Close     5746 non-null   float64       
 5   Volume    5746 non-null   int64         
 6   Currency  5746 non-null   object        
dtypes: datetime64[ns](1), float64(4), int64(1), object(1)
memory usage: 314.4+ KB
In [4]:
# 'Close' is the target variable
# drop 'Currency' column
df.drop(['Currency'], axis=1, inplace=True)
In [5]:
df.describe()
Out[5]:
Open High Low Close Volume
count 5746.000000 5746.000000 5746.000000 5746.000000 5746.000000
mean 127.267635 128.847034 125.784669 127.215567 8807.178907
std 50.569425 51.164948 49.851487 50.506519 9612.789034
min 41.500000 42.000000 41.500000 41.500000 0.000000
25% 98.800000 100.112500 97.862500 98.650000 61.000000
50% 120.400000 121.600000 118.950000 120.250000 7008.000000
75% 144.800000 146.000000 143.000000 144.337500 14497.750000
max 305.300000 306.250000 304.000000 304.900000 62045.000000
In [6]:
# Resample the data on 'Close' price weekly
sale_weekly = df.resample('W', on='Date').mean()

# Resample the data on 'Close' price monthly
sale_monthly = df.resample('M', on='Date').mean()

# Visualization
fig, (ax1, ax2, ax3) = plt.subplots(3, figsize=(12,8))

sns.lineplot(x='Date', y='Close', data=df, ax=ax1)
ax1.set_title('Coffee Daily Close Price')

sns.lineplot(x='Date', y='Close', data=sale_weekly, ax=ax2)
ax2.set_title('Coffee Weekly Close Price')

sns.lineplot(x='Date', y='Close', data=sale_monthly, ax=ax3)
ax3.set_title('Coffee Monthly Close Price')

plt.tight_layout()

Train the model¶

In [7]:
# Dataset for forecasting
coffee_price = df[['Date','Close']]
coffee_price = coffee_price.rename(columns = {'Date':'ds', 'Close':'y'})

# Splitting train-test sets
train_set = coffee_price[:-365]
test_set = coffee_price[-365:]

# Fit the model to the train data
m = Prophet(yearly_seasonality=True )
m.add_country_holidays(country_name='US')
m.fit(train_set)

18:19:40 - cmdstanpy - INFO - Chain [1] start processing
18:19:46 - cmdstanpy - INFO - Chain [1] done processing
Out[7]:
<prophet.forecaster.Prophet at 0x1bb86685370>

Forecast¶

In [8]:
# Forecast model
prophet_pred = m.predict(test_set)

# Plot the forecast results
pred_plot = m.plot(prophet_pred)
sns.scatterplot(x='ds', y='y', data=test_set, color = 'g', label='True Daily close price')
Out[8]:
<AxesSubplot:xlabel='ds', ylabel='y'>

Findings¶

  • The forecast for the daily close price of coffee over a period of 365 days (solid blue line) is far off from the true close price (green line).
  • Indicating that the data can be influenced by other features leading to a very low accuracy prediction.

Further Analysis¶

Deeper look into the features resulted from the Test set¶

In [9]:
prophet_pred.head()
Out[9]:
ds trend yhat_lower yhat_upper trend_lower trend_upper Christmas Day Christmas Day_lower Christmas Day_upper Christmas Day (Observed) ... weekly weekly_lower weekly_upper yearly yearly_lower yearly_upper multiplicative_terms multiplicative_terms_lower multiplicative_terms_upper yhat
0 2021-03-25 101.894239 87.358103 119.462934 101.894239 101.894239 0.0 0.0 0.0 0.0 ... -0.884969 -0.884969 -0.884969 2.089961 2.089961 2.089961 0.0 0.0 0.0 103.099231
1 2021-03-26 101.874364 87.590018 119.470158 101.874364 101.874364 0.0 0.0 0.0 0.0 ... -1.077511 -1.077511 -1.077511 1.926802 1.926802 1.926802 0.0 0.0 0.0 102.723655
2 2021-03-29 101.814739 87.273833 116.874786 101.814739 101.814739 0.0 0.0 0.0 0.0 ... -1.095343 -1.095343 -1.095343 1.541970 1.541970 1.541970 0.0 0.0 0.0 102.261366
3 2021-03-30 101.794863 86.100350 118.323247 101.794863 101.794863 0.0 0.0 0.0 0.0 ... -0.865276 -0.865276 -0.865276 1.453662 1.453662 1.453662 0.0 0.0 0.0 102.383249
4 2021-03-31 101.774988 87.045318 117.011452 101.774988 101.774988 0.0 0.0 0.0 0.0 ... -0.829819 -0.829819 -0.829819 1.386737 1.386737 1.386737 0.0 0.0 0.0 102.331906

5 rows × 70 columns

In [10]:
# plotting on the predicted values from test set
prophet_plot = prophet_pred[['ds','trend','yearly','holidays','yhat']]
prophet_plot = prophet_plot.rename(columns={'yearly':'season'})
prophet_plot.head()
Out[10]:
ds trend season holidays yhat
0 2021-03-25 101.894239 2.089961 0.0 103.099231
1 2021-03-26 101.874364 1.926802 0.0 102.723655
2 2021-03-29 101.814739 1.541970 0.0 102.261366
3 2021-03-30 101.794863 1.453662 0.0 102.383249
4 2021-03-31 101.774988 1.386737 0.0 102.331906
In [11]:
prophet_plot = prophet_plot.sort_values(by='ds')
fig = make_subplots(rows=3, cols=1, subplot_titles=('trend','season','holidays'))

fig.add_trace(go.Scatter(x=prophet_plot['ds'], y=prophet_plot['trend']), row=1, col=1)
fig.add_trace(go.Scatter(x=prophet_plot['ds'], y=prophet_plot['season']), row=2, col=1)
fig.add_trace(go.Scatter(x=prophet_plot['ds'], y=prophet_plot['holidays']), row=3, col=1)
In [12]:
# Fit the model to the data
m = Prophet(yearly_seasonality=True)
m.add_country_holidays(country_name='US')
m.fit(coffee_price)

18:19:47 - cmdstanpy - INFO - Chain [1] start processing
18:19:53 - cmdstanpy - INFO - Chain [1] done processing
Out[12]:
<prophet.forecaster.Prophet at 0x1bb8b1c4bb0>
In [13]:
# Predict for the next 365-day period
df_pred = m.make_future_dataframe(periods=365, include_history=True)
coffee_pred = m.predict(df_pred)
In [14]:
coffee_plot = coffee_pred[['ds','trend','yearly','holidays']]
coffee_plot = coffee_plot.rename(columns={'yearly':'season'})
coffee_plot.head()
Out[14]:
ds trend season holidays
0 2000-01-03 106.606151 3.499964 0.0
1 2000-01-04 106.509236 3.641218 0.0
2 2000-01-05 106.412322 3.763193 0.0
3 2000-01-06 106.315408 3.865800 0.0
4 2000-01-07 106.218494 3.949306 0.0
In [15]:
coffee_plot = coffee_plot.sort_values(by='ds')
fig = make_subplots(rows=3, cols=1, subplot_titles=('trend','season','holidays'))

fig.add_trace(go.Scatter(x=coffee_plot['ds'], y=coffee_plot['trend']), row=1, col=1)
fig.add_trace(go.Scatter(x=coffee_plot['ds'], y=coffee_plot['season']), row=2, col=1)
fig.add_trace(go.Scatter(x=coffee_plot['ds'], y=coffee_plot['holidays']), row=3, col=1)

fig.show()

Breaking down trend and seasonality¶

For the test set

In [16]:
m.plot_components(prophet_pred)
Out[16]:

For the original set

In [17]:
m.plot_components(coffee_pred)
Out[17]:

Conclusion¶

  • According to the data we have analyzed, it appears that the coffee stock price did not have a clear trend over the period of 2000-2020. While coffee is a widely consumed product, with demand expected to stay stable or increase, we did see a sharp increase in coffee prices starting in 2011 and reaching its peak in 2012. This was likely due to poor weather and farming conditions that resulted in a poor harvest.[1]

  • However, the model did not accurately predict coffee prices for the period of 2021-2022. This can be explained by unexpected factors such as poor weather conditions and supply chain issues in Brazil, one of the largest coffee producers in the world, as well as the impact of COVID-19 on freight costs.

  • It is also worth noting that coffee prices typically drop in May and pick up again in late July, as seen when analyzing the data for both 2022 and a hypothetical extra 365-day period in 2023. This seasonal trend was observed in both:
    • The comparison of the predicted coffee prices for 2022 with the actual prices in 2022.
    • The prediction of coffee prices for an additional 365-day period (equivalent to one year) in 2023, based on historical data.



[1] https://www.theguardian.com/business/2011/apr/21/commodities-coffee-shortage-price-rise-expected

In [ ]: